It starts with a mess of thoughts and ends with an articulated story: how AI helps me refine my thinking and could help you with yours
Yes, I use AI to help me write this Substack. Here is how I do it:
AI prompts me with a question: “What happened in the past two or three weeks that has stayed with you most?” I then start with a thought that seems important and begin typing. AI reacts to what I write by asking me questions, one at a time, probing my thinking and feelings about an event or comment. The conversation sometimes diverges, suggesting that my thoughts are drawn towards something else more important. We go back and forth like this for quite some time, sometimes 30–40 minutes.
During this time, I find that my thinking about a problem or event becomes clearer. This is not surprising: writing is an integral part of understanding. In one study, students doing laboratory writing assignments improved their critical-thinking scores, particularly in analysis and inference, while their peers doing quiz-based labs did not show the same improvement. I remember that during my undergraduate physics degree in Padova, if I wanted to get a good grade in a course, I had better handwrite my notes. For me, typing was not as helpful as handwriting, but it was better than not writing at all.
When I am satisfied with the clarity of my thinking, I ask AI to produce a skeleton draft, which I then read sentence by sentence and rewrite or edit as appropriate.
The conversation as an educational tool
When AI prompted me today, what stayed with me most from the past week was an interview with Eric Bradlow on the Love Conquers Fear podcast. Bradlow is a professor of statistics and data science and Wharton’s Vice Dean of AI & Analytics. One part especially resonated. He said he no longer writes a prompt sheet to get a task done. He tells AI what he wants to achieve and asks it to question him until he is ready for it to build something. “I see these conversations as an educational tool for me,” he said.
I recognised that process immediately. It is close to what I have been doing with these articles: using questions and answers to find out what I think before asking AI to produce a draft. His phrase “educational tool” opened my eyes to the possibility that AI could help us become smarter.
Bradlow went on to talk about the impact of AI in academia, and his statements gave me goosebumps.
He says “academia is about to face a huge problem”. Universities have traditionally attracted talented people through prestige and the promise of interesting work. AI could make it easier for companies to build their own expertise and keep their most interesting data and problems in-house. “Most professors are stuck in 1995,” he says, provocatively. “Who’s our customers? The students! Most students have more experience and know more about using LLM than professors.”
The Wharton AI & Analytics Initiative, led by Bradlow, offers students experiential learning: they learn methods and then apply them to real problems. Students need to be challenged with problems that do not have an answer at the back of the book. They need to make independent decisions, explain their reasoning, fail safely, learn from failure and try again. AI should be part of that experience because it will also be part of many workplaces.
Bradlow asks, “How do I educate my peers? How do I bring professors along in this AI journey?” That question has been on my mind, too.
We need to experiment with how we use AI to educate ourselves
We are far from finding the best way for people to work with AI. We are still discovering what happens when we change the relationship: let the model question us, give it a specific role in a project, or let several people shape what it learns from. We should ask ourselves whether those interactions help us create something new, understand something more deeply or become more capable without the model.
For example, artist Sougwen Chung draws alongside a robotic arm trained on her earlier drawings. Her past gestures influence what the machine draws; the strokes of the AI-controlled arm give her something unexpected to respond to. This exchange may push her to explore new shapes and designs rather than asking a machine to make or finish a drawing for her.
© Sougwen Chung / SCILICET
I was also inspired by The Call, made by Holly Herndon and Mat Dryhurst with community choirs. They wrote material for people to sing, recorded fifteen choirs across the UK and used the recordings to create choral AI models. The project explores how choristers can have a say in the use of models trained on their voices. A New Scientist review, quoted by the Serpentine, says that the work “tackles some thorny questions about AI and creativity – and stirs the soul with music”. I love this line. It captures the tension between artificial and human creation while acknowledging that the result can touch the soul.
There is an example even closer to my own research. In a 2021 Nature paper, mathematicians and AI researchers, including Demis Hassabis, used machine learning to find mathematical patterns that guided their intuition towards new conjectures and theorems. Although the paper appeared early in the current wave of AI research, I find it inspiring that the model “nudged” the researchers to see a connection hidden among large clouds of numbers. Human thinking then turned a pattern-recognition exercise into a formal contribution to mathematics. By the way, we are very close to figuring out how LLMs understand knot topology (just in case Demis is reading) — but that is a whole other story.
None of these examples gives us a ready-made model for improving students’ experiences and learning. They do make me want to experiment with how we can use AI to help people learn. Humans are naturally curious, and that curiosity has helped us make sense of the world for a very long time. AI could become a companion for curious people: it can help us explore many topics, whenever we like, with extraordinary patience.
In some ways, I am already benefiting from my experiment. I have never been a prolific writer, but I am enjoying posting on Substack through this new process! Perhaps we should find a way to use AI to help students (re)discover the joy of learning, not because they have to pass an exam, but because they are curious and passionate about their topic and want to know more.
An invitation to co-create
Inspired by all this, I have started experimenting with conversational tools that could help students reason through problems, refine their thinking and perhaps make learning more enjoyable and attuned to their needs. It is early work. Before I take it much further, I want to hear how people actually learn and what happens when they bring AI into that process.
Poll: How could AI help you learn more effectively and enjoyably?
- Explaining a difficult concept in different ways until one makes sense to me.
- Explaining a difficult concept by referring and elaborating on the lecture notes used in class.
- Asking questions and giving feedback as I explain what I am learning in my own words.
- Helping me apply a method to a real problem or project, for example, by creating problems for me to try to solve.
- Something else.
- I do not see AI as a useful educational tool.
And please tell me in the comments: How do you learn best? Where does using AI help you, and where does it get in the way? I would especially like to hear from students, PhD researchers, postdocs and teachers. If you think AI has no place in education, tell me why. I need to hear that too!
Until the next time,
Davide
Join the conversation
How do you learn best? Where does AI help you, and where does it get in the way? I would love to hear your experience.
Send me your thoughts by email
Your reply will come directly to me by email.